Yeah, I think the problem here is we have 2 different definitions of "safe." If there is an option to punish, it's unsafe. Like doval said, if they know what is coming and know their options and cannot punish, it is safe. Similarily, safe on shield means if they shield it and know their options, they cannot punish.
You said "safe against most characters" so I think you understand that you cannot just blanket call this move safe.
problem is... we're using the same definition of 'safe' and coming to different conclusions.
Empty SH is always safe with some distance because at any time the opponent does anything, you can react to it. I don't want to go too far down the well here, but if you are far enough away that opponent has to move to reach you, rather than just throwing an attack (which even then, you can react to), it's safe. It's not like by jumping, we're set to an arc where we can't act or adjust, so there's no justification for calling that unsafe or vulnerable. (even being high in the air isn't necessarily considered a bad position for MM, given his good aerial movement with rush jumping, etc)
And yes, I've carried that "against most characters" pretty consistently because there are exceptions to rules. However, because this is a "mega man's weaknesses" thread, I'm using as close to "average" conditions as I can get, so if a move is safe/good/whathaveyou against MOST of the cast but DK stomps it for whatever reason... that shouldn't really count as a weakness, instead of a matchup-specific thing. An inverse of this is something like the leaf shield footstool edgeguard. It's REALLY good against a small number of chars, so it just gets lumped into "edgeguarding options" for me, because you can't ever say it's a universally good idea.
So let me try once more for the bair discussion using actual examples. first off, I've been shorthanding a bit that could have led to some confusion, so I will go more in depth this time. Bair does have a disjoint at the edge, and all my posting is assuming good spacing by using the tip of the disjoint. The opponent's shield is a circle around them, so contact with the shield occurs at a larger radius than contact with the opponent. What this does for my discussion is essentially guarantee that the upper and lower slashes of bair connect with the shield, even if only the middle slash would hit the opponent. It also adds range to the shield pressure of the bair that wouldn't actually hit the opponent, but I'm mostly going to neglect that.
Assuming good timing and spacing, bair connects with the shield directly before landing, at the tip of its range. At this point, the opponent has ~20 frames of time to react and punish (for the sake of sanity, I am assuming that the time between when the bair hits and MM lands to be roughly equal to the shield stun that the hit inflicts on the opponent). The opponent has a minimum of 7 frames of lag to drop shield after the attack and punish. MM is currently outside of opponent grab range of anybody without a tether (tether doesn't have time to come out, either), and is about the range away of a decent sized smash attack.
Here's where my shorthand has come in. At this point, most chars A) don't have a single attack to reach MM without a dash first [which takes too long] or B) have an attack that will reach, but is too slow to come out at that range in 13 frames.
Example A) Lucario. He drops shield and can't run fast enough to reach MM, and doesn't have a reaching forward move (other than fsmash or force palm) to reach MM, and both of those are too slow to connect, so MM doesn't get punished. (there are quite a few other chars in this group, who have fast moves or ranged moves, but not both)
Example B) Ike, Ganon, Bowser, etc. Have a plethora of moves that would reach, but nothing fast enough to connect
Example C) Mario, Doc, Luigi, Peach, etc. Don't have moves that reach MM without moving. Can't move in time.
All that said, I KNOW THERE ARE EXCEPTIONS here. For example, I
believe Samus can drop shield ftilt to punish... but again those are (imo) exceptions and not the rule for the "average" matchup.